ChatGPT o3 mini | New AI Model Disaster For OpenAI

In this video, we take a hard look at OpenAI’s latest release, ChatGPT o3 mini and why it might be more of a disaster than a breakthrough. Despite the hype, o3-mini appears to be only marginally better than its predecessor, offering few substantial improvements at a time when competitors like DeepSeek are circling.

[@roynijland7386]

o3-mini with THE RIGHT prompts at the RIGHT timing in the right context is just the key. The LLM`s are usually smart enough, but its just us humans that are too dumb to ask it in the exact right way haha. I like using both o3-mini, deepseek-r1 and claude sonnet. This combo is just unbeatable, and with ai agents in the mix add gpt-4o-mini

[@Arekadiusz]

Can you tell exactly for what do you use each of them?

[@spiph23]

This video reminds me of when Android companies first got their stuff together and many people were saying how Apple was screwed because the iPhone was so inferior to whatever Android phone was out. Personally, I think OpenAI are doing pretty good and that calling it a disaster is, perhaps, a touch hyperbolic.

[@Pabz2030]

o3 mini is certainly quick. But it doesnt seem to follow instructions particularly well. I prefer using a local LLM like phi-4 or mistral 24B for nearly everything these days. If I were prepared to pay $200 per month for a significant access to a decent online model I'd definitely be prepared to save that up for a decent home LLM server instead that can comfortably run a local 70B model.

[@clarazegarelli5861]

Interesting, but curious how do you update it? can learn new things I mean new training? is it connected to internet to search for real time responses?

[@lwwells]

O3 mini high is wonderful for python

[@josephgoodwin3244]

I like using both o3 and 4o

[@voltax4435]

True, sonnet still the king at coding tasks.

[@brianginn1993]

Refreshing take

[@Daab4]

Has it happened to anyone else that o3 does its thinking process in some random foreign language?

[@TrainTheAlgo]

Haven't seen that yet 🫤

[@TobiasGeek]

yep for me it was chinese

[@fkxfkx]

english?